Databases
Redis, ClickHouse, PostgreSQL, OpenSearch -- internals, clustering, HA
10 lessons
[PROD] ClickHouse Production Architecture
Complete production-ready ClickHouse with Keeper, proper scaling, failure scenarios, and best practices
[S3] ClickHouse over S3
ClickHouse with S3-backed SharedMergeTree: disaggregated storage, zero-copy replication, local SSD cache, background merges on S3
[HOT/COLD] ClickHouse + Iceberg
HOT/COLD tiered storage: 7-day HOT in ClickHouse (3x, SSD, ~11ms) + 90-day COLD in Apache Iceberg/S3 (Parquet, ~200ms), 45% cost savings
[MULTI-DC] ClickHouse Geo-Distributed
ClickHouse across 3 datacenters (US/EU/AP) with 5-node Keeper, sync/async replication, and DR scenarios
[PROD] OpenSearch Complete Architecture
Production OpenSearch with vector search, ML nodes, hot/warm tiers, sharding, and 8 scenarios
[FIXED] OpenSearch Correct Architecture
Fixed: All clients through LB, flexible shard replication, all nodes know all masters
Redis Cluster + Sentinel
Redis Cluster with 3 shards (master + replica), Sentinel failover, write/read/failover scenarios
[HA] Redis Sentinel (NO Sharding)
Redis Sentinel: Single master + replicas with automatic failover, client discovery, SDOWN→ODOWN consensus, split-brain prevention
[CLUSTER] Redis Cluster (Native Sharding)
Redis Cluster: 16384 hash slots, built-in failover via gossip, MOVED/ASK redirects, online resharding - NO Sentinel needed
[DEEP] PostgreSQL + Patroni HA
Production HA: Patroni + etcd consensus, streaming replication (sync/async), HAProxy failover, WAL archiving